Problems of Classification by Pseudoinversion Means: Feature Space Transformation, Influence of Initial Information Perturbations

摘要

For the problem of information hyperplane classification by means of matrix pseudoinversion there are offered the technologies for space optimization of feature vectors. The estimates of perturbation influence of feature space vectors components on conditions of classes separation and the separation strip width are obtained. The conducted research is based on results of the perturbation theory of pseudoinverse and projective matrices.